Questions on `integrator.success_iter` vs. `integrator.destats.naccept`

[ranocha]

Both integrator.success_iter and integrator.destats.naccept seem to contain the number of accepted steps/iterations - as far as I understand from their names and implementations. However, there is a subtly but important difference when trying to use them in, say, custom callbacks.

• integrator.success_iter is updated in loopheader!, see https://github.com/SciML/OrdinaryDiffEq.jl/blob/cdc5ef291d37f31ce7746e9f27aeeeae5d806393/src/integrators/integrator_utils.jl#L10
• integrator.destats.naccept is updated in loopheader!, see https://github.com/SciML/OrdinaryDiffEq.jl/blob/cdc5ef291d37f31ce7746e9f27aeeeae5d806393/src/integrators/integrator_utils.jl#L192 and https://github.com/SciML/OrdinaryDiffEq.jl/blob/cdc5ef291d37f31ce7746e9f27aeeeae5d806393/src/integrators/integrator_utils.jl#L209

Thus, integrator.success_iter == integrator.destats.naccept - 1 at the end of the integration and whenever callbacks are called. If all steps are accepted, integrator.success_iter == integrator.iter - 1.

My questions are

• Is there a reason for this different behavior of integrator.success_iter and integrator.destats.naccept?
• If not: Can we get rid of integrator.success_iter or change it's behavior such that it mimics integrator.destats.naccept?

[ChrisRackauckas]

success_iter is just needed by a few methods. It probably can be eliminated though.